[en] In this paper, two issues in the FSS iteration method, i.e., the waiting time for surface source data and the variance biases in local tallies are investigated for the domain decomposed, 3-D continuous-energy whole-core calculation. The fission sources are provided as usual, while the surface sources are provided by banking MC particles crossing local domain boundaries. The surface sources serve as boundary conditions for nonoverlapping local problems, so that each local problem can be solved independently. In this paper, two issues in the FSS iteration are investigated. One is quantifying the waiting time of processors to receive surface source data. By using nonblocking communication, 'time penalty' to wait for the arrival of the surface source data is reduced. The other important issue is underestimation of the sample variance of the tally because of additional inter-iteration correlations in surface sources. From the numerical results on a 3-D whole-core test problem, it is observed that the time penalty is negligible in the FSS iteration method and that the real variances of both pin powers and assembly powers are estimated by the HB method. For those purposes, three cases; Case 1 (1 local domain), Case 2 (4 local domains), Case 3 (16 local domains) are tested. For both Cases 2 and 3, the time penalties for waiting are negligible compared to the source-tracking times. However, for finer divisions of local domains, the loss of parallel efficiency caused by the different number of sources for local domains in symmetric locations becomes larger due to the stochastic errors in source distributions. For all test cases, the HB method very well estimates the real variances of local tallies. However, it is also noted that the real variances of local tallies estimated by the HB method show slightly smaller than the real variances obtained from 30 independent batch runs and the deviations become larger for finer divisions of local domains. The batch size used for the HB method may not be large enough to accurately estimate the real variance in this test problem. In conclusion, the waiting time for surface source data is negligible in the FSS iteration method, and the loss of parallel efficiency caused by the different number of sources can be reduced by the source splitting scheme discussed.